Nhungamiro yehunyanzvi

Hinge Loss

Hinge kurasikirwa kunoranga muenzaniso wechikamu kana muganho wayo wakasainwa ukawira pasi peimwe, kusanganisira nemapoinzi akaiswa zvakanaka anogara padyo nemuganho wesarudzo.

Teerera kugwaro iriAI-yakarondedzerwa · ~3 min

Pa peji ino4 min verenga
  1. Pfupiso
  2. Kudzika Kwakadzika
  3. Kuitwa muNyika yeChokwadi
  4. Ramangwana reHinge Loss
  5. Kuzvishandisa
  6. Ramba Uchiongorora
  7. Mibvunzo inowanzo bvunzwa

Pfupiso

Icho chiri pakati pekutsigira vector michina nekuti kuderedza kurasikirwa kwehinge padivi pechirango chemuganho kunokurudzira muganho unopatsanura unobvumira zvimwe zvikanganiso uchitsvaga muganho wakafara.

Kudzika Kwakadzika

Pakusarudzwa kwemabhinari, nyora mavara anoti y mu {-1, +1} uye rega f(x) ive yesarudzo yesarudzo. Mucheto wakasainwa ndi y nguva f(x). Hinge kurasikirwa ndeye max(0, 1 - y f(x)). Muenzaniso wakasarudzika une zvibodzwa munzira kwayo unogona kuramba uchirasikirwa kana uri mukati mechikamu chechikamu. Poindi pamusoro kana kupfuura pamucheto ine zero hinge kurasikirwa. Nzvimbo yakasarudzika ine muchero wakasainwa wakashata uye inowana chirango chinodarika chimwe. Kurasikirwa uku kunokurudzira mugadziri kuti aparadzanise makirasi ane muganho, kwete kungopa iwo iwo chaiwo mavara ekudzidzisa. Mune mutsara wetsigiro vector muchina, chinangwa chinoyera hupamhi hwemaginji kutarisana nehuwandu hwekurasikirwa kwehinge, ine yemazuva ese paramende inodzora tradeoff. Chirango chakasimba chekutyora chinogona kukwana data yekudzidziswa zvakasimba; kubvumira kutyorwa kwakawanda kunogona kufarira muganhu wakakura, wakapfava. Kernel nzira dzinoshandura chinomiririra ichichengeta iyo margin-yakavakirwa chinangwa. Funga nezvemuenzaniso wekufungidzira une label +1 uye zvibodzwa 0.4. Muganho wayo ndeye 0.4 uye kurasikirwa kwayo kuri 0.6. Kana mumwe muenzaniso wakanaka une zvibodzwa 1.3, kurasikirwa kwayo i zero. Kana nyaya ine negative-yakanyorwa iine zvibodzwa 0.5, muganho wayo wakasainwa i -0.5 uye kurasikirwa kwayo kwakaenzana ne1.5. Aya maitiro anoratidza kuti nei kurongeka kwega kusingataridzi kurasikirwa kwehinge. Logistic kurasikirwa kune zvakare convex uye margin-sensitive, asi kusiyana nekurasikirwa kwehinge inoenderera ichipa chirango chakanaka kumienzaniso yakarongedzerwa, kunyangwe kure kumucheto. Iyo probabilistic fomu inobatanidza zvakasikwa kune log-inogona. Hinge kurasikirwa haisiriyo pachayo fungidziro yezvingangoita fungidziro. MaSVM anogona kupetwa pamwe nekukwanisa kuenzanisa kana zvingangove zvichidikanwa, uchishandisa nzira yakakodzera yekusimbisa. Chikero maficha, tune kudzoreredza pamapeta ekudzidzisa, uye ongorora mutengo wekukanganisa pane mumiriri data. Kurasikirwa kwehinge hakutsananguri sarudzo yakanaka yesarudzo yechikumbiro chese; mhedzisiro yebhizinesi inogona kuda chikumbaridzo chakasiyana kana kirasi-yekuremedza sarudzo.

Kuitwa muNyika yeChokwadi

Yekufungidzira bhinari classifier inopa kesi yakanyorwa nemazvo mupendero wakasainwa we 0.4. Kurasika kwayo hinge kuri max(0, 1 - 0.4) = 0.6, nekuti poindi iri kudivi chairo asi mukati memucheto.

Muenzaniso wakanyorwa zvisirizvo wakasaina muganho -0.5. Kurasika kwayo hinge kuri max(0, 1 - (-0.5)) = 1.5, ichiratidza divi rayo risiri iro uye chinhambwe kubva painodiwa.

Inotsigira-vector classifier inofananidza kurasikirwa kwehinge nekurasikirwa kwezvinhu panguva yekudzidziswa. Chinangwa cheSVM chinosimbisa kutyorwa kwemargin; Logistic kurasikirwa inoramba ichipindura kumienzaniso yese uye inopa dudziro inogoneka chete nekuwedzera modhi kana kuenzanisa.

Injiniya anoyera maficha ekuisa asati agadzirisa iyo SVM's dhizaini paramende. Zviyero zvezvikero zvinokanganisa kureba uye miganho, saka kugadzirisa pasina kuenderana preprocessing kunogona kuvhiringidza kuenzanisa.

Ramangwana reHinge Loss

Margin-based classification inoramba ichibatsira kana muganho wakajeka uye kutonga pamusoro pekutyorwa kwemargin zvine basa. Inoshanda workflows inofanirwa kutaura kana iyo yekushandisa inoshandisa hinge kana yakapetwa hinge, kuti maficha akayerwa sei, uye kuti kukosha kwekugara kwakasarudzwa sei. Kana vashandisi vepasi vachida fungidziro dzenjodzi, zvikwata zvinofanirwa kuongorora zvingangoitika zvakasiyana pane kushandura zvibodzwa zvesarudzo zvakananga muzvikamu. Kuongorora kunogona kuenzanisa kugoverwa kwemargin uye kirasi-chaiyo yekukanganisa mitengo nekufamba kwenguva. Zvirinani zvinotsanangudza zvinogona kuratidza kuti ndezvipi zvakaonekwa zviri pedyo nemuganhu, apo kusavimbika kunowanzonyanya kukosha, ukuwo uchichengeta zvibodzwa zvakasiyana kubva pane zvingangoitika.

Kuzvishandisa

Nei zvichikosha

  • Mutengo uye bhajeti. Zvisarudzo zvezvivakwa zvinotyaira kuita uye mutengo wekushandisa kwemakore.
  • Sarudzo dzakajeka. Dzidzo yehunyanzvi inobatsira zvikwata kusarudza murwi wakakodzera, kwete iwo mutsva chete.
  • Kudzora kwemhando yepamusoro. Sarudzo dzeinjiniya dziri nani dzinoderedza zviitiko zvekuvimbika mukugadzira.

Chenjera

  • Kugadzirisa imwe bhenji kunogona kuvanza yakafara system kushaya simba.
  • Infrastructure uye mari yekugadzirisa inowanzotarisirwa pasi.
  • Chengetedzo uye kucherechedzwa mapundu anogona kukura sezvo masisitimu anowedzera kuoma.

Matanho anotevera

  1. Tsanangura latency, mhando, uye mutengo zvinangwa usati waitwa.
  2. Benchmark pasi pechokwadi mutoro uye data mamiriro.
  3. Chishandiso chekutarisa zvikanganiso, kudonha, uye mushandisi maitiro.
  4. Gadzirira nzira dzekudzosera kumashure uye dzezviitiko usati wawedzera.

Ramba Uchiongorora

Tsamba yemahara

Tora iyo yemazuva ese AI pfupiso

Nhatu dzakasimbiswa dzeAI nyaya mangwanani ega ega pakati pevhiki, dzakanyorwa neChirungu chakajeka. Yemahara zvachose, hapana ads.

Imwe email zuva rega rega revhiki. Bvisa kunyoresa kamwe chete. Hatimbotengese kana kugovera kero yako.

Zviedze iwe pachako

Take the Hinge Loss quiz

Mhinduro yepakarepo pamhinduro yega yega, uye chitupa chinogovaniswa chine ID inobatika kana wapedza kosi.

Tanga mibvunzo

Tsigira dzidzo yemahara yeAI. AI Understanding ndeye 501(c)(3) isingabatsiri - hapana ads, hapana paywall, zvachose. Ita chipo

Mibvunzo inowanzo bvunzwa

Chii chinonzi Hinge Loss?

Hinge kurasikirwa kunoranga muenzaniso wechikamu kana muganho wayo wakasainwa ukawira pasi peimwe, kusanganisira nemapoinzi akaiswa zvakanaka anogara padyo nemuganho wesarudzo. Icho chiri pakati pekutsigira vector michina nekuti kuderedza kurasikirwa kwehinge padivi pechirango chemuganho kunokurudzira muganho unopatsanura unobvumira zvimwe zvikanganiso uchitsvaga muganho wakafara.

Ndepapi pakasainwa muganho apo kurasikirwa kwehinge kunotanga kuve zero?

Kana mupendero wakasainwa uchingoita imwe chete, 1 minus haina kunaka uye iyo yakakwira ne zero i zero.

Mhosva yakanyorwa zvisina kunaka inogamuchira zvibodzwa 0.5. Ndeipi muganho wayo wakasainwa?

Mucheto wakasainwa iregi nguva zvibodzwa; zvakaipa imwe nguva 0.5 inokwana -0.5.

Sei nzvimbo yakanyatsorongedzerwa ichigona kuve neyakanaka hinge kurasikirwa?

Hinge kurasikirwa kunoranga mapoinzi ane akasainwa margin pazasi peimwe, kunyangwe divi rakafanotaurwa riri rechokwadi.

Kurasika kwezvinhu kunosiyana sei kune yakanyatsorongedzerwa nzvimbo iri kure nemuganho?

Kurasikirwa kwezvinhu kunodzikira asi kunoramba kuri kwakanaka kune zvibodzwa zvinogumira, nepo kurasikirwa kwehinge kuri zero kupfuura margin.

Ndeipi chinangwa tradeoff chakajairika mumutsara SVM?

Chinangwa cheSVM chinoenzanisa yakajairika-yakavakirwa margin nguva uye kurangwa hinge-kurasikirwa kutyora.